With the rapid development of computer technology, the man-machine communication's technique use more and more widespread. Speech recognition technology makes man-machine communication more nature and humanization. Speech recognition refers to the technology that computer by identifying and understanding converts the natural voice signals into texts, speech recognition technology adopt principles of pattern matching, choose the optimum matching reference pattern for recognitory result by comparison unknown and known sound pattern. The feature extraction is the very important part in speech recognition technology, we can extract feature parameter from speech by MFCC and IPCC, and it will affect recognitory result.Elementary selection is first step in speech recognition technology; speech recognition technology elementary has word, syllable, and phoneme. Syllable is phonetic ultimate unit, and it can be classified into single syllable and multisyllable. In Tibetan there are 30 consonants and 5 vowels constitute 289 syllables. there is the word()consist of one syllable at least, and word()consist of seven syllables at most in Tibetan, each syllable component with consonant, vowel and tone, and separate by syllabic sign.This thesis main research on the Tibetan voice feature parameter based on specific people small vocabulary from digital sampling, noise filter, dewing and framing, studied the wave of time domain, the wave off frequency domain and the analysis of sonogram. A1gorithm with two restricted values threshold are applied to endpoint detection of Ando Tibetan speech signals. Analyze the predictive cepstral coefficients (LPCC), Mel Frequency Cepstrum Coefficient (MFCC).
|